1
/* Innobase relational database engine; Copyright (C) 2001 Innobase Oy
3
This program is free software; you can redistribute it and/or modify
4
it under the terms of the GNU General Public License 2
5
as published by the Free Software Foundation in June 1991.
7
This program is distributed in the hope that it will be useful,
8
but WITHOUT ANY WARRANTY; without even the implied warranty of
9
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
10
GNU General Public License for more details.
12
You should have received a copy of the GNU General Public License 2
13
along with this program (in file COPYING); if not, write to the Free
14
Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */
15
/******************************************************
16
The database buffer pool high-level routines
20
Created 11/5/1995 Heikki Tuuri
21
*******************************************************/
28
#include "mtr0types.h"
29
#include "buf0types.h"
31
#include "hash0hash.h"
35
/* Flags for flush types */
36
#define BUF_FLUSH_LRU 1
37
#define BUF_FLUSH_SINGLE_PAGE 2
38
#define BUF_FLUSH_LIST 3 /* An array in the pool struct
39
has size BUF_FLUSH_LIST + 1: if you
40
add more flush types, put them in
42
/* Modes for buf_page_get_gen */
43
#define BUF_GET 10 /* get always */
44
#define BUF_GET_IF_IN_POOL 11 /* get if in pool */
45
#define BUF_GET_NOWAIT 12 /* get if can set the latch without
47
#define BUF_GET_NO_LATCH 14 /* get and bufferfix, but set no latch;
48
we have separated this case, because
49
it is error-prone programming not to
50
set a latch, and it should be used
52
/* Modes for buf_page_get_known_nowait */
53
#define BUF_MAKE_YOUNG 51
54
#define BUF_KEEP_OLD 52
55
/* Magic value to use instead of checksums when they are disabled */
56
#define BUF_NO_CHECKSUM_MAGIC 0xDEADBEEFUL
58
extern buf_pool_t* buf_pool; /* The buffer pool of the database */
60
extern ibool buf_debug_prints;/* If this is set TRUE, the program
61
prints info whenever read or flush
63
#endif /* UNIV_DEBUG */
64
extern ulint srv_buf_pool_write_requests; /* variable to count write request
67
/************************************************************************
68
Creates the buffer pool. */
73
/* out, own: buf_pool object, NULL if not
74
enough memory or error */
75
ulint max_size, /* in: maximum size of the buf_pool in
77
ulint curr_size, /* in: current size to use, must be <=
78
max_size, currently must be equal to
80
ulint n_frames); /* in: number of frames; if AWE is used,
81
this is the size of the address space window
82
where physical memory pages are mapped; if
83
AWE is not used then this must be the same
85
/*************************************************************************
86
Gets the current size of buffer buf_pool in bytes. In the case of AWE, the
87
size of AWE window (= the frames). */
90
buf_pool_get_curr_size(void);
91
/*========================*/
92
/* out: size in bytes */
93
/*************************************************************************
94
Gets the maximum size of buffer pool in bytes. In the case of AWE, the
95
size of AWE window (= the frames). */
98
buf_pool_get_max_size(void);
99
/*=======================*/
100
/* out: size in bytes */
101
/************************************************************************
102
Gets the smallest oldest_modification lsn for any page in the pool. Returns
103
ut_dulint_zero if all modified pages have been flushed to disk. */
106
buf_pool_get_oldest_modification(void);
107
/*==================================*/
108
/* out: oldest modification in pool,
109
ut_dulint_zero if none */
110
/*************************************************************************
111
Allocates a buffer frame. */
114
buf_frame_alloc(void);
115
/*==================*/
116
/* out: buffer frame */
117
/*************************************************************************
118
Frees a buffer frame which does not contain a file page. */
123
buf_frame_t* frame); /* in: buffer frame */
124
/*************************************************************************
125
Copies contents of a buffer frame to a given buffer. */
131
byte* buf, /* in: buffer to copy to */
132
buf_frame_t* frame); /* in: buffer frame */
133
/******************************************************************
134
NOTE! The following macros should be used instead of buf_page_get_gen,
135
to improve debugging. Only values RW_S_LATCH and RW_X_LATCH are allowed
137
#define buf_page_get(SP, OF, LA, MTR) buf_page_get_gen(\
139
BUF_GET, __FILE__, __LINE__, MTR)
140
/******************************************************************
141
Use these macros to bufferfix a page with no latching. Remember not to
142
read the contents of the page unless you know it is safe. Do not modify
143
the contents of the page! We have separated this case, because it is
144
error-prone programming not to set a latch, and it should be used
146
#define buf_page_get_with_no_latch(SP, OF, MTR) buf_page_get_gen(\
147
SP, OF, RW_NO_LATCH, NULL,\
148
BUF_GET_NO_LATCH, __FILE__, __LINE__, MTR)
149
/******************************************************************
150
NOTE! The following macros should be used instead of buf_page_get_gen, to
151
improve debugging. Only values RW_S_LATCH and RW_X_LATCH are allowed as LA! */
152
#define buf_page_get_nowait(SP, OF, LA, MTR) buf_page_get_gen(\
154
BUF_GET_NOWAIT, __FILE__, __LINE__, MTR)
155
/******************************************************************
156
NOTE! The following macros should be used instead of
157
buf_page_optimistic_get_func, to improve debugging. Only values RW_S_LATCH and
158
RW_X_LATCH are allowed as LA! */
159
#define buf_page_optimistic_get(LA, BL, G, MC, MTR) \
160
buf_page_optimistic_get_func(LA, BL, G, MC, __FILE__, __LINE__, MTR)
161
/************************************************************************
162
This is the general function used to get optimistic access to a database
166
buf_page_optimistic_get_func(
167
/*=========================*/
168
/* out: TRUE if success */
169
ulint rw_latch,/* in: RW_S_LATCH, RW_X_LATCH */
170
buf_block_t* block, /* in: guessed block */
171
buf_frame_t* guess, /* in: guessed frame; note that AWE may move
173
dulint modify_clock,/* in: modify clock value if mode is
174
..._GUESS_ON_CLOCK */
175
const char* file, /* in: file name */
176
ulint line, /* in: line where called */
177
mtr_t* mtr); /* in: mini-transaction */
178
/************************************************************************
179
Tries to get the page, but if file io is required, releases all latches
180
in mtr down to the given savepoint. If io is required, this function
181
retrieves the page to buffer buf_pool, but does not bufferfix it or latch
185
buf_page_get_release_on_io(
186
/*=======================*/
187
/* out: pointer to the frame, or NULL
188
if not in buffer buf_pool */
189
ulint space, /* in: space id */
190
ulint offset, /* in: offset of the page within space
191
in units of a page */
192
buf_frame_t* guess, /* in: guessed frame or NULL */
193
ulint rw_latch, /* in: RW_X_LATCH, RW_S_LATCH,
195
ulint savepoint, /* in: mtr savepoint */
196
mtr_t* mtr); /* in: mtr */
197
/************************************************************************
198
This is used to get access to a known database page, when no waiting can be
202
buf_page_get_known_nowait(
203
/*======================*/
204
/* out: TRUE if success */
205
ulint rw_latch,/* in: RW_S_LATCH, RW_X_LATCH */
206
buf_frame_t* guess, /* in: the known page frame */
207
ulint mode, /* in: BUF_MAKE_YOUNG or BUF_KEEP_OLD */
208
const char* file, /* in: file name */
209
ulint line, /* in: line where called */
210
mtr_t* mtr); /* in: mini-transaction */
211
/************************************************************************
212
This is the general function used to get access to a database page. */
217
/* out: pointer to the frame or NULL */
218
ulint space, /* in: space id */
219
ulint offset, /* in: page number */
220
ulint rw_latch,/* in: RW_S_LATCH, RW_X_LATCH, RW_NO_LATCH */
221
buf_frame_t* guess, /* in: guessed frame or NULL */
222
ulint mode, /* in: BUF_GET, BUF_GET_IF_IN_POOL,
224
const char* file, /* in: file name */
225
ulint line, /* in: line where called */
226
mtr_t* mtr); /* in: mini-transaction */
227
/************************************************************************
228
Initializes a page to the buffer buf_pool. The page is usually not read
229
from a file even if it cannot be found in the buffer buf_pool. This is one
230
of the functions which perform to a block a state transition NOT_USED =>
231
FILE_PAGE (the other is buf_page_init_for_read above). */
236
/* out: pointer to the frame, page bufferfixed */
237
ulint space, /* in: space id */
238
ulint offset, /* in: offset of the page within space in units of
240
mtr_t* mtr); /* in: mini-transaction handle */
241
/************************************************************************
242
Inits a page to the buffer buf_pool, for use in ibbackup --restore. */
245
buf_page_init_for_backup_restore(
246
/*=============================*/
247
ulint space, /* in: space id */
248
ulint offset, /* in: offset of the page within space
249
in units of a page */
250
buf_block_t* block); /* in: block to init */
251
/************************************************************************
252
Decrements the bufferfix count of a buffer control block and releases
253
a latch, if specified. */
258
buf_block_t* block, /* in: buffer block */
259
ulint rw_latch, /* in: RW_S_LATCH, RW_X_LATCH,
261
mtr_t* mtr); /* in: mtr */
262
/************************************************************************
263
Moves a page to the start of the buffer pool LRU list. This high-level
264
function can be used to prevent an important page from from slipping out of
270
buf_frame_t* frame); /* in: buffer frame of a file page */
271
/************************************************************************
272
Returns TRUE if the page can be found in the buffer pool hash table. NOTE
273
that it is possible that the page is not yet read from disk, though. */
278
/* out: TRUE if found from page hash table,
279
NOTE that the page is not necessarily yet read
281
ulint space, /* in: space id */
282
ulint offset);/* in: page number */
283
/************************************************************************
284
Returns the buffer control block if the page can be found in the buffer
285
pool. NOTE that it is possible that the page is not yet read
286
from disk, though. This is a very low-level function: use with care! */
291
/* out: control block if found from page hash table,
292
otherwise NULL; NOTE that the page is not necessarily
293
yet read from disk! */
294
ulint space, /* in: space id */
295
ulint offset);/* in: page number */
296
/************************************************************************
297
Resets the check_index_page_at_flush field of a page if found in the buffer
301
buf_reset_check_index_page_at_flush(
302
/*================================*/
303
ulint space, /* in: space id */
304
ulint offset);/* in: page number */
305
/************************************************************************
306
Sets file_page_was_freed TRUE if the page is found in the buffer pool.
307
This function should be called when we free a file page and want the
308
debug version to check that it is not accessed any more unless
312
buf_page_set_file_page_was_freed(
313
/*=============================*/
314
/* out: control block if found from page hash table,
316
ulint space, /* in: space id */
317
ulint offset); /* in: page number */
318
/************************************************************************
319
Sets file_page_was_freed FALSE if the page is found in the buffer pool.
320
This function should be called when we free a file page and want the
321
debug version to check that it is not accessed any more unless
325
buf_page_reset_file_page_was_freed(
326
/*===============================*/
327
/* out: control block if found from page hash table,
329
ulint space, /* in: space id */
330
ulint offset); /* in: page number */
331
/************************************************************************
332
Recommends a move of a block to the start of the LRU list if there is danger
333
of dropping from the buffer pool. NOTE: does not reserve the buffer pool
337
buf_block_peek_if_too_old(
338
/*======================*/
339
/* out: TRUE if should be made younger */
340
buf_block_t* block); /* in: block to make younger */
341
/************************************************************************
342
Returns the current state of is_hashed of a page. FALSE if the page is
343
not in the pool. NOTE that this operation does not fix the page in the
344
pool if it is found there. */
347
buf_page_peek_if_search_hashed(
348
/*===========================*/
349
/* out: TRUE if page hash index is built in search
351
ulint space, /* in: space id */
352
ulint offset);/* in: page number */
353
/************************************************************************
354
Gets the youngest modification log sequence number for a frame.
355
Returns zero if not file page or no modification occurred yet. */
358
buf_frame_get_newest_modification(
359
/*==============================*/
360
/* out: newest modification to page */
361
buf_frame_t* frame); /* in: pointer to a frame */
362
/************************************************************************
363
Increments the modify clock of a frame by 1. The caller must (1) own the
364
pool mutex and block bufferfix count has to be zero, (2) or own an x-lock
368
buf_frame_modify_clock_inc(
369
/*=======================*/
371
buf_frame_t* frame); /* in: pointer to a frame */
372
/************************************************************************
373
Increments the modify clock of a frame by 1. The caller must (1) own the
374
buf_pool mutex and block bufferfix count has to be zero, (2) or own an x-lock
378
buf_block_modify_clock_inc(
379
/*=======================*/
381
buf_block_t* block); /* in: block */
382
/************************************************************************
383
Returns the value of the modify clock. The caller must have an s-lock
384
or x-lock on the block. */
387
buf_block_get_modify_clock(
388
/*=======================*/
390
buf_block_t* block); /* in: block */
391
/************************************************************************
392
Calculates a page checksum which is stored to the page when it is written
393
to a file. Note that we must be careful to calculate the same value
394
on 32-bit and 64-bit architectures. */
397
buf_calc_page_new_checksum(
398
/*=======================*/
400
byte* page); /* in: buffer page */
401
/************************************************************************
402
In versions < 4.0.14 and < 4.1.1 there was a bug that the checksum only
403
looked at the first few bytes of the page. This calculates that old
405
NOTE: we must first store the new formula checksum to
406
FIL_PAGE_SPACE_OR_CHKSUM before calculating and storing this old checksum
407
because this takes that field as an input! */
410
buf_calc_page_old_checksum(
411
/*=======================*/
413
byte* page); /* in: buffer page */
414
/************************************************************************
415
Checks if a page is corrupt. */
418
buf_page_is_corrupted(
419
/*==================*/
420
/* out: TRUE if corrupted */
421
byte* read_buf); /* in: a database page */
422
/**************************************************************************
423
Gets the page number of a pointer pointing within a buffer frame containing
427
buf_frame_get_page_no(
428
/*==================*/
429
/* out: page number */
430
byte* ptr); /* in: pointer to within a buffer frame */
431
/**************************************************************************
432
Gets the space id of a pointer pointing within a buffer frame containing a
436
buf_frame_get_space_id(
437
/*===================*/
439
byte* ptr); /* in: pointer to within a buffer frame */
440
/**************************************************************************
441
Gets the space id, page offset, and byte offset within page of a
442
pointer pointing to a buffer frame containing a file page. */
445
buf_ptr_get_fsp_addr(
446
/*=================*/
447
byte* ptr, /* in: pointer to a buffer frame */
448
ulint* space, /* out: space id */
449
fil_addr_t* addr); /* out: page offset and byte offset */
450
/**************************************************************************
451
Gets the hash value of the page the pointer is pointing to. This can be used
452
in searches in the lock hash table. */
455
buf_frame_get_lock_hash_val(
456
/*========================*/
457
/* out: lock hash value */
458
byte* ptr); /* in: pointer to within a buffer frame */
459
/**************************************************************************
460
Gets the mutex number protecting the page record lock hash chain in the lock
467
byte* ptr); /* in: pointer to within a buffer frame */
468
/***********************************************************************
469
Gets the frame the pointer is pointing to. */
474
/* out: pointer to frame */
475
byte* ptr); /* in: pointer to a frame */
476
/***********************************************************************
477
Checks if a pointer points to the block array of the buffer pool (blocks, not
483
/* out: TRUE if pointer to block */
484
void* ptr); /* in: pointer to memory */
486
/*************************************************************************
487
Validates the buffer pool data structure. */
492
/*************************************************************************
493
Prints info of the buffer pool data structure. */
498
#endif /* UNIV_DEBUG */
499
/************************************************************************
500
Prints a page to stderr. */
505
byte* read_buf); /* in: a database page */
506
/*************************************************************************
507
Returns the number of latched pages in the buffer pool. */
510
buf_get_latched_pages_number(void);
511
/*==============================*/
512
/*************************************************************************
513
Returns the number of pending buf pool ios. */
516
buf_get_n_pending_ios(void);
517
/*=======================*/
518
/*************************************************************************
519
Prints info of the buffer i/o. */
524
FILE* file); /* in: file where to print */
525
/*************************************************************************
526
Returns the ratio in percents of modified pages in the buffer pool /
527
database pages in the buffer pool. */
530
buf_get_modified_ratio_pct(void);
531
/*============================*/
532
/**************************************************************************
533
Refreshes the statistics used to print per-second averages. */
536
buf_refresh_io_stats(void);
537
/*======================*/
538
/*************************************************************************
539
Checks that all file pages in the buffer are in a replaceable state. */
544
/*************************************************************************
545
Checks that there currently are no pending i/o-operations for the buffer
549
buf_pool_check_no_pending_io(void);
550
/*==============================*/
551
/* out: TRUE if there is no pending i/o */
552
/*************************************************************************
553
Invalidates the file pages in the buffer pool when an archive recovery is
554
completed. All the file pages buffered must be in a replaceable state when
555
this function is called: not latched and not modified. */
558
buf_pool_invalidate(void);
559
/*=====================*/
561
/*========================================================================
562
--------------------------- LOWER LEVEL ROUTINES -------------------------
563
=========================================================================*/
565
/************************************************************************
566
Maps the page of block to a frame, if not mapped yet. Unmaps some page
567
from the end of the awe_LRU_free_mapped. */
570
buf_awe_map_page_to_frame(
571
/*======================*/
572
buf_block_t* block, /* in: block whose page should be
574
ibool add_to_mapped_list);/* in: TRUE if we in the case
575
we need to map the page should also
577
awe_LRU_free_mapped list */
578
#ifdef UNIV_SYNC_DEBUG
579
/*************************************************************************
580
Adds latch level info for the rw-lock protecting the buffer frame. This
581
should be called in the debug version after a successful latching of a
582
page if we know the latching order level of the acquired latch. */
585
buf_page_dbg_add_level(
586
/*===================*/
587
buf_frame_t* frame, /* in: buffer page where we have acquired
589
ulint level); /* in: latching order level */
590
#endif /* UNIV_SYNC_DEBUG */
591
/*************************************************************************
592
Gets a pointer to the memory frame of a block. */
597
/* out: pointer to the frame */
598
buf_block_t* block); /* in: pointer to the control block */
599
/*************************************************************************
600
Gets the space id of a block. */
606
buf_block_t* block); /* in: pointer to the control block */
607
/*************************************************************************
608
Gets the page number of a block. */
611
buf_block_get_page_no(
612
/*==================*/
613
/* out: page number */
614
buf_block_t* block); /* in: pointer to the control block */
615
/***********************************************************************
616
Gets the block to whose frame the pointer is pointing to. */
621
/* out: pointer to block */
622
byte* ptr); /* in: pointer to a frame */
623
/************************************************************************
624
This function is used to get info if there is an io operation
625
going on on a buffer page. */
630
/* out: TRUE if io going on */
631
buf_block_t* block); /* in: pool block, must be bufferfixed */
632
/***********************************************************************
633
Accessor function for block array. */
636
buf_pool_get_nth_block(
637
/*===================*/
638
/* out: pointer to block */
639
buf_pool_t* pool, /* in: pool */
640
ulint i); /* in: index of the block */
641
/************************************************************************
642
Function which inits a page for read to the buffer buf_pool. If the page is
643
(1) already in buf_pool, or
644
(2) if we specify to read only ibuf pages and the page is not an ibuf page, or
645
(3) if the space is deleted or being deleted,
646
then this function does nothing.
647
Sets the io_fix flag to BUF_IO_READ and sets a non-recursive exclusive lock
648
on the buffer frame. The io-handler must take care that the flag is cleared
649
and the lock released later. This is one of the functions which perform the
650
state transition NOT_USED => FILE_PAGE to a block (the other is
654
buf_page_init_for_read(
655
/*===================*/
656
/* out: pointer to the block or NULL */
657
ulint* err, /* out: DB_SUCCESS or DB_TABLESPACE_DELETED */
658
ulint mode, /* in: BUF_READ_IBUF_PAGES_ONLY, ... */
659
ulint space, /* in: space id */
660
ib_longlong tablespace_version,/* in: prevents reading from a wrong
661
version of the tablespace in case we have done
663
ulint offset);/* in: page number */
664
/************************************************************************
665
Completes an asynchronous read or write request of a file page to or from
669
buf_page_io_complete(
670
/*=================*/
671
buf_block_t* block); /* in: pointer to the block in question */
672
/************************************************************************
673
Calculates a folded value of a file page address to use in the page hash
677
buf_page_address_fold(
678
/*==================*/
679
/* out: the folded value */
680
ulint space, /* in: space id */
681
ulint offset);/* in: offset of the page within space */
682
/**********************************************************************
683
Returns the control block of a file page, NULL if not found. */
688
/* out: block, NULL if not found */
689
ulint space, /* in: space id */
690
ulint offset);/* in: offset of the page within space */
691
/***********************************************************************
692
Increments the pool clock by one and returns its new value. Remember that
693
in the 32 bit version the clock wraps around at 4 billion! */
696
buf_pool_clock_tic(void);
697
/*====================*/
698
/* out: new clock value */
699
/*************************************************************************
700
Gets the current length of the free list of buffer blocks. */
703
buf_get_free_list_len(void);
704
/*=======================*/
708
/* The buffer control block structure */
710
struct buf_block_struct{
712
/* 1. General fields */
714
ulint magic_n; /* magic number to check */
715
ulint state; /* state of the control block:
716
BUF_BLOCK_NOT_USED, ...; changing
717
this is only allowed when a thread
718
has BOTH the buffer pool mutex AND
719
block->mutex locked */
720
byte* frame; /* pointer to buffer frame which
721
is of size UNIV_PAGE_SIZE, and
722
aligned to an address divisible by
723
UNIV_PAGE_SIZE; if AWE is used, this
724
will be NULL for the pages which are
725
currently not mapped into the virtual
726
address space window of the buffer
728
os_awe_t* awe_info; /* if AWE is used, then an array of
730
UNIV_PAGE_SIZE / OS_AWE_X86_PAGE_SIZE
731
(normally = 4) physical memory
732
pages; otherwise NULL */
733
ulint space; /* space id of the page */
734
ulint offset; /* page number within the space */
735
ulint lock_hash_val; /* hashed value of the page address
736
in the record lock hash table */
737
mutex_t mutex; /* mutex protecting this block:
738
state (also protected by the buffer
739
pool mutex), io_fix, buf_fix_count,
740
and accessed; we introduce this new
741
mutex in InnoDB-5.1 to relieve
742
contention on the buffer pool mutex */
743
rw_lock_t lock; /* read-write lock of the buffer
745
buf_block_t* hash; /* node used in chaining to the page
747
ibool check_index_page_at_flush;
748
/* TRUE if we know that this is
749
an index page, and want the database
750
to check its consistency before flush;
751
note that there may be pages in the
752
buffer pool which are index pages,
753
but this flag is not set because
754
we do not keep track of all pages */
755
/* 2. Page flushing fields */
757
UT_LIST_NODE_T(buf_block_t) flush_list;
758
/* node of the modified, not yet
759
flushed blocks list */
760
dulint newest_modification;
761
/* log sequence number of the youngest
762
modification to this block, zero if
764
dulint oldest_modification;
765
/* log sequence number of the START of
766
the log entry written of the oldest
767
modification to this block which has
768
not yet been flushed on disk; zero if
769
all modifications are on disk */
770
ulint flush_type; /* if this block is currently being
771
flushed to disk, this tells the
772
flush_type: BUF_FLUSH_LRU or
775
/* 3. LRU replacement algorithm fields */
777
UT_LIST_NODE_T(buf_block_t) free;
778
/* node of the free block list */
779
ibool in_free_list; /* TRUE if in the free list; used in
781
UT_LIST_NODE_T(buf_block_t) LRU;
782
/* node of the LRU list */
783
UT_LIST_NODE_T(buf_block_t) awe_LRU_free_mapped;
784
/* in the AWE version node in the
785
list of free and LRU blocks which are
787
ibool in_LRU_list; /* TRUE of the page is in the LRU list;
789
ulint LRU_position; /* value which monotonically
790
decreases (or may stay constant if
791
the block is in the old blocks) toward
792
the end of the LRU list, if the pool
793
ulint_clock has not wrapped around:
794
NOTE that this value can only be used
795
in heuristic algorithms, because of
796
the possibility of a wrap-around! */
797
ulint freed_page_clock;/* the value of freed_page_clock
798
of the buffer pool when this block was
799
the last time put to the head of the
800
LRU list; a thread is allowed to
801
read this for heuristic purposes
802
without holding any mutex or latch */
803
ibool old; /* TRUE if the block is in the old
804
blocks in the LRU list */
805
ibool accessed; /* TRUE if the page has been accessed
806
while in the buffer pool: read-ahead
807
may read in pages which have not been
808
accessed yet; this is protected by
809
block->mutex; a thread is allowed to
810
read this for heuristic purposes
811
without holding any mutex or latch */
812
ulint buf_fix_count; /* count of how manyfold this block
813
is currently bufferfixed; this is
814
protected by block->mutex */
815
ulint io_fix; /* if a read is pending to the frame,
816
io_fix is BUF_IO_READ, in the case
817
of a write BUF_IO_WRITE, otherwise 0;
818
this is protected by block->mutex */
819
/* 4. Optimistic search field */
821
dulint modify_clock; /* this clock is incremented every
822
time a pointer to a record on the
823
page may become obsolete; this is
824
used in the optimistic cursor
825
positioning: if the modify clock has
826
not changed, we know that the pointer
827
is still valid; this field may be
828
changed if the thread (1) owns the
829
pool mutex and the page is not
830
bufferfixed, or (2) the thread has an
831
x-latch on the block */
833
/* 5. Hash search fields: NOTE that the first 4 fields are NOT
834
protected by any semaphore! */
836
ulint n_hash_helps; /* counter which controls building
837
of a new hash index for the page */
838
ulint n_fields; /* recommended prefix length for hash
839
search: number of full fields */
840
ulint n_bytes; /* recommended prefix: number of bytes
841
in an incomplete field */
842
ibool left_side; /* TRUE or FALSE, depending on
843
whether the leftmost record of several
844
records with the same prefix should be
845
indexed in the hash index */
847
/* These 6 fields may only be modified when we have
848
an x-latch on btr_search_latch AND
849
a) we are holding an s-latch or x-latch on block->lock or
850
b) we know that block->buf_fix_count == 0.
852
An exception to this is when we init or create a page
853
in the buffer pool in buf0buf.c. */
855
ibool is_hashed; /* TRUE if hash index has already been
856
built on this page; note that it does
857
not guarantee that the index is
858
complete, though: there may have been
859
hash collisions, record deletions,
861
ulint n_pointers; /* used in debugging: the number of
862
pointers in the adaptive hash index
863
pointing to this frame */
864
ulint curr_n_fields; /* prefix length for hash indexing:
865
number of full fields */
866
ulint curr_n_bytes; /* number of bytes in hash indexing */
867
ibool curr_left_side; /* TRUE or FALSE in hash indexing */
868
dict_index_t* index; /* Index for which the adaptive
869
hash index has been created. */
870
/* 6. Debug fields */
871
#ifdef UNIV_SYNC_DEBUG
872
rw_lock_t debug_latch; /* in the debug version, each thread
873
which bufferfixes the block acquires
874
an s-latch here; so we can use the
875
debug utilities in sync0rw */
877
ibool file_page_was_freed;
878
/* this is set to TRUE when fsp
879
frees a page in buffer pool */
882
#define BUF_BLOCK_MAGIC_N 41526563
884
/* The buffer pool structure. NOTE! The definition appears here only for
885
other modules of this directory (buf) to see it. Do not use from outside! */
887
struct buf_pool_struct{
889
/* 1. General fields */
891
mutex_t mutex; /* mutex protecting the buffer pool
892
struct and control blocks, except the
893
read-write lock in them */
894
byte* frame_mem; /* pointer to the memory area which
895
was allocated for the frames; in AWE
896
this is the virtual address space
897
window where we map pages stored
898
in physical memory */
899
byte* frame_zero; /* pointer to the first buffer frame:
900
this may differ from frame_mem, because
901
this is aligned by the frame size */
902
byte* high_end; /* pointer to the end of the buffer
904
ulint n_frames; /* number of frames */
905
buf_block_t* blocks; /* array of buffer control blocks */
906
buf_block_t** blocks_of_frames;/* inverse mapping which can be used
907
to retrieve the buffer control block
908
of a frame; this is an array which
909
lists the blocks of frames in the
911
frame_zero + UNIV_PAGE_SIZE, ...
912
a control block is always assigned
913
for each frame, even if the frame does
914
not contain any data; note that in AWE
915
there are more control blocks than
917
os_awe_t* awe_info; /* if AWE is used, AWE info for the
918
physical 4 kB memory pages associated
919
with buffer frames */
920
ulint max_size; /* number of control blocks ==
921
maximum pool size in pages */
922
ulint curr_size; /* current pool size in pages;
923
currently always the same as
925
hash_table_t* page_hash; /* hash table of the file pages */
927
ulint n_pend_reads; /* number of pending read operations */
929
time_t last_printout_time; /* when buf_print was last time
931
ulint n_pages_read; /* number read operations */
932
ulint n_pages_written;/* number write operations */
933
ulint n_pages_created;/* number of pages created in the pool
935
ulint n_page_gets; /* number of page gets performed;
936
also successful searches through
937
the adaptive hash index are
938
counted as page gets; this field
939
is NOT protected by the buffer
941
ulint n_pages_awe_remapped; /* if AWE is enabled, the
942
number of remaps of blocks to
944
ulint n_page_gets_old;/* n_page_gets when buf_print was
945
last time called: used to calculate
947
ulint n_pages_read_old;/* n_pages_read when buf_print was
949
ulint n_pages_written_old;/* number write operations */
950
ulint n_pages_created_old;/* number of pages created in
951
the pool with no read */
952
ulint n_pages_awe_remapped_old;
953
/* 2. Page flushing algorithm fields */
955
UT_LIST_BASE_NODE_T(buf_block_t) flush_list;
956
/* base node of the modified block
958
ibool init_flush[BUF_FLUSH_LIST + 1];
959
/* this is TRUE when a flush of the
960
given type is being initialized */
961
ulint n_flush[BUF_FLUSH_LIST + 1];
962
/* this is the number of pending
963
writes in the given flush type */
964
os_event_t no_flush[BUF_FLUSH_LIST + 1];
965
/* this is in the set state when there
966
is no flush batch of the given type
968
ulint ulint_clock; /* a sequence number used to count
969
time. NOTE! This counter wraps
970
around at 4 billion (if ulint ==
972
ulint freed_page_clock;/* a sequence number used to count the
973
number of buffer blocks removed from
974
the end of the LRU list; NOTE that
975
this counter may wrap around at 4
976
billion! A thread is allowed to
977
read this for heuristic purposes
978
without holding any mutex or latch */
979
ulint LRU_flush_ended;/* when an LRU flush ends for a page,
980
this is incremented by one; this is
981
set to zero when a buffer block is
984
/* 3. LRU replacement algorithm fields */
986
UT_LIST_BASE_NODE_T(buf_block_t) free;
987
/* base node of the free block list;
988
in the case of AWE, at the start are
989
always free blocks for which the
990
physical memory is mapped to a frame */
991
UT_LIST_BASE_NODE_T(buf_block_t) LRU;
992
/* base node of the LRU list */
993
buf_block_t* LRU_old; /* pointer to the about 3/8 oldest
994
blocks in the LRU list; NULL if LRU
995
length less than BUF_LRU_OLD_MIN_LEN */
996
ulint LRU_old_len; /* length of the LRU list from
997
the block to which LRU_old points
998
onward, including that block;
999
see buf0lru.c for the restrictions
1000
on this value; not defined if
1002
UT_LIST_BASE_NODE_T(buf_block_t) awe_LRU_free_mapped;
1003
/* list of those blocks which are
1004
in the LRU list or the free list, and
1005
where the page is mapped to a frame;
1006
thus, frames allocated, e.g., to the
1007
locki table, are not in this list */
1010
/* States of a control block */
1011
#define BUF_BLOCK_NOT_USED 211 /* is in the free list */
1012
#define BUF_BLOCK_READY_FOR_USE 212 /* when buf_get_free_block returns
1013
a block, it is in this state */
1014
#define BUF_BLOCK_FILE_PAGE 213 /* contains a buffered file page */
1015
#define BUF_BLOCK_MEMORY 214 /* contains some main memory object */
1016
#define BUF_BLOCK_REMOVE_HASH 215 /* hash index should be removed
1017
before putting to the free list */
1019
/* Io_fix states of a control block; these must be != 0 */
1020
#define BUF_IO_READ 561
1021
#define BUF_IO_WRITE 562
1023
/************************************************************************
1024
Let us list the consistency conditions for different control block states.
1026
NOT_USED: is in free list, not in LRU list, not in flush list, nor
1028
READY_FOR_USE: is not in free list, LRU list, or flush list, nor page
1030
MEMORY: is not in free list, LRU list, or flush list, nor page
1032
FILE_PAGE: space and offset are defined, is in page hash table
1033
if io_fix == BUF_IO_WRITE,
1034
pool: no_flush[block->flush_type] is in reset state,
1035
pool: n_flush[block->flush_type] > 0
1037
(1) if buf_fix_count == 0, then
1038
is in LRU list, not in free list
1040
if and only if oldest_modification > 0
1042
if and only if io_fix == BUF_IO_READ
1044
if and only if io_fix == BUF_IO_WRITE
1046
(2) if buf_fix_count > 0, then
1047
is not in LRU list, not in free list
1049
if and only if oldest_modification > 0
1050
if io_fix == BUF_IO_READ,
1052
if io_fix == BUF_IO_WRITE,
1057
NOT_USED => READY_FOR_USE
1058
READY_FOR_USE => MEMORY
1059
READY_FOR_USE => FILE_PAGE
1061
FILE_PAGE => NOT_USED NOTE: This transition is allowed if and only if
1062
(1) buf_fix_count == 0,
1063
(2) oldest_modification == 0, and
1068
#include "buf0buf.ic"
added
removed
storage/innobase/include/buf0buf.h
